Searching for Nested Oscillations in Frequency and Sensor Space. Will Penny. Wellcome Trust Centre for Neuroimaging. University College London.

Transcription

1 in Frequency and Sensor Space Oscillation Wellcome Trust Centre for Neuroimaging. University College London. Non- Workshop on Non-Invasive Imaging of Nonlinear Interactions. 20th Annual Computational Neuroscience (CNS) Meeting. 28th July 2011, Stockholm

2 Oscillation Oscillation Non- Phase Amplitude Coupling (PAC).

3 Canolty et al (2006) define the modulation index as 1 N M = z[n] N where n=1 z[n] = a γ [n] exp (iφ θ [n]) The significance of M is then assessed using a surrogate data approach. Oscillation Non-

4 Vanhatalo et al. (2004) and Mormann et al. (2005) use the (PLV) between the phase of the lower frequency oscillation and the phase of the amplitude envelope of the higher frequency oscillation 1 PLV = N N exp ( i(φ θ [n] φ aγ [n]) ) n=1 The significance of PLV is then assessed using a surrogate data approach. Oscillation Non-

5 Oscillation Bruns and Eckhorn (2004) define the as ESC = Corr(x θ [n], a γ [n]) The significance of ESC is assessed using t distributions. Non-

6 Penny et al. (2008) use a (GLM) approach based on the multiple regression model a γ = Xβ + e where β are regression coefficients, e is additive Gaussian noise and the design matrix X contains three columns: cos(φ θ [n]) sin(φ θ [n]) A column of 1 s Oscillation Non- Significance is assessed using F-tests over the first two regression coefficients. More generally, X could be a Fourier series.

7 ECoG Data Data from Kai Miller and Jeff Ojemann at Washington State. They collected from subjects performing a one-back visual working memory task. Each item (picture of a house) was presented twice. On the second presentation of the item subjects press a button. On the second presentation the item therefore does nt need to be remembered. On the first presentation it does. We computed PAC measures for each trial between 6Hz theta and high gamma (76 to 200Hz). Oscillation Non- The measures were then Gaussianised for each trial, and we tested for between condition (remember vs not) differences using two sample t-tests at each electrode.

8 ESC and GLM detect nested oscillations that the other measures don t. ECoG Data ESC (top left), GLM (top right), PLV (bottom left), (bottom right). Oscillation Non-

9 ECoG Data Oscillation Non- Current item does not need to be remembered.

10 ECoG Data Oscillation Non- Current item needs to be remembered.

11 A population of Slow GABA-A cells inhibits a population of Fast GABA-A cells. Oscillation Non- Each cell is a single compartment Hodgkin-Huxley model (White et al, 1998).

12 Populations of GABA-B (top,slow) and GABA-A (bottom,fast) cells. Oscillation Non- Fast cells had synaptic rise times of 1ms and fall times of 9ms. For the slow cells they are 5ms and 150ms.

13 Comparison of PAC measures. Oscillation Non- GLM (green), PLV (black), ESC (red), (blue). See Penny et al. (2008) for many further tests.

14 Experimental Paradigm Oscillation Non- of Visual Working Memory (Fuentemilla et al. 2010).

15 Multivariate Analysis at Encoding Oscillation Non- Multivariate classification based on sensor space spectra using features from 13 to 80 Hz.

16 Multivariate Classification of Maintenance Oscillation Non- Greater replay during memory conditions.

17 Replay is Phase-Locked to Theta Theta activity was then projected to source space, and for each source, Poch et al. (2011) computed the phases at which patterns were replayed. To see if these phases were non-uniformly distributed, a PLV measure was computed for each source. Poch et al. (2011) then tested to see which sources had PLVs that predicted of memory performance. This identified a right hippocampal and a right inferior frontal region. Oscillation Non-

18 Theta Sources Oscillation Non-

19 Theta Sources Oscillation Non-

20 Processing Stream Extract phase of theta activity in source region. Extract time-frequency maps at each sensor, v, from frequencies f = 16 : 4 : 128 Hz during delay period. For each trial compute GLM PAC measure. Record fitted regression coefficients s fv and c fv. The sine and cosine terms for each frequency and sensor Create a NIFTI format image for each measure. There are 3 conditions and 40 trials per condition, with 2 measures per trial. This gives 240 data points per subject Set up design matrix in SPM and implement a GLM analysis in space-frequency Litvak et al, 2010) Use Random Field Theory to correct for multiple comparisons Oscillation Non-

21 Images are entered in the following order Sine coefficients for Sine coefficients for Non-Config Sine coefficients for Config Cos coefficients for Cos coefficients for Non-Config Cos coefficients for Config Oscillation Non-

22 Oscillation Non-

23 The statistical signifiance of phase amplitude coupling is corrected for the multiple comparisons over space and frequency using Random Field Theory. Oscillation Non- We can use the standard threshold eg FWE=0.05.

24 Non- Oscillation Non-

25 Non- Oscillation Non-

26 Non- Oscillation Non-

27 Oscillation Non-

28 Oscillation Non-

29 Oscillation Non-

30 G. Buzsaki (2006) Rhythms of the Brain. Oxford University Press. R. Canolty et al (2006) High gamma power is phase-locked to theta oscillations in neocortex. Science 313, L. Fuentemilla et al (2010) Theta-coupled periodic replay in working memory. Current Biology 20, 1-7. V. Litvak et al. (2011) EEG and MEG data analysis in SPM8. Comput Intell Neurosci. Article ID: K. Miller et al. (2009) Power-Law Scaling in the Brain Surface Electric Potential. PLoS CB, 5(12):e F. Mormann et al. (2005) Phase/amplitude reset and theta-gamma interaction in the human medial temporal lobe. Hippocampus 15: W. Penny et al (2008) Testing for Oscillation. Journal of Neuroscience Methods, 174, C. Poch et al (2011) theta-phase modulation of replay correlates with configural-relational short-term memory performance. Journal of Neuroscience, 31(19): S. Vanhatalo et al. (2004) Infraslow oscillations modulate excitability and interictal activity in the human cortex during sleep. PNAS 101(14): Oscillation Non- J. White et al. (2000) s of interneurons with fast and slow GABA-A kinetics. Proc Natl Acad Sci USA, 97(14):